Pith. sign in

REVIEW 3 major objections 3 minor 17 references

Monte Carlo methods on compact symplectic manifolds

T0 review · 3 major / 3 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read On prequantized compact symplectic manifolds, a point process from Bochner–Schrödinger spectral projections yields an unbiased, asymptotically normal Monte Carlo estimator at the optimal worst-case rate.

desk verdict A genuine extension of DPP quadrature to prequantized symplectic manifolds with a clean CLT, but the central variance estimate is deferred to the author's to-appear companion paper. read the letter →

arxiv 2608.07021 v1 pith:ZVW2DDP4 submitted 2026-08-07 math.DG cs.NAmath-phmath.MPmath.NAmath.PR

Yuri A. Kordyukov This is my paper · ORCID
classification math.DGcs.NAmath-phmath.MPmath.NAmath.PR MSC 53D5060G5565C0558J50
keywords MonteCarlointegrationdeterminantalpointprocessescompactsymplecticmanifoldsBochner–SchrödingeroperatorLandaulevelscentrallimittheoremoptimalworst-caserateprequantization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Prequantized compact symplectic manifolds carry a natural operator $H_p=\frac{1}{p}\Delta^{L^p}+V$ on sections of $L^p$, and the paper makes this operator the engine of numerical integration. It defines a determinantal point process from the spectral projection onto a window $I$ that avoids the set of Landau levels, draws quadrature nodes from it, and proves that the estimator $\widehat J=\sum_i f(x_i)/P_{p,I}(x_i,x_i)$ is unbiased. The main result is a central limit theorem: the rescaled error converges to a centered normal law with the variance displayed in (1.13). Consequently the mean squared error decays as $\sigma^2/N_p^{(n+1)/n}$, the optimal worst-case rate for randomized $C^1$ integration in dimension $2n$, now realized on curved, compact symplectic spaces rather than only in Euclidean space.

What carries the argument

The engine is the determinantal point process associated with the finite-rank spectral projection $P_{p,I}$ of $H_p=\frac{1}{p}\Delta^{L^p}+V$, restricted to an interval $I=(\alpha,\beta)$ that avoids the Landau-level set $\Sigma$. Its $N_p$-point law is the squared Slater determinant of an orthonormal basis of the spectral subspace $\mathcal H_p$, and the proof tracks the log-Laplace transform $F_p(t)=-\log\mathbb E[e^{-t\Xi_p}]$. The second derivative of $F_p$ reduces to a Hilbert–Schmidt norm of the commutator $[P_{p,I,t},f_p]$ with respect to tilted inner products, and known kernel asymptotics for $P_{p,I}(x,y)$ convert this into the explicit variance (1.13). The off-diagonal exponential decay of the kernel, inherited from the spectral gap, is what makes the commutator norm concentrate on small distances and the central limit hold.

What would settle it

Take a compact prequantized symplectic manifold satisfying the gap assumption, for instance a two-torus with $g=g_B$ and $V=0$, simulate the determinantal point process at increasing $p$, and check whether the variance of $\widehat J$ follows $\sigma^2/N_p^{(n+1)/n}$ with the constant from (1.13); a mismatch would disprove the central limit theorem.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 1.4: for any $f\in C^1(X,\mathbb R)$, the random variable $\Xi_p$ defined in (1.12) converges in distribution, as $p\to\infty$, to $N(0,\sigma^2)$, where $\sigma^2$ is given by (1.13) in terms of the squared gradient of $f_B$ with respect to the metric-like quantity $|\cdot|_I$ built from the chosen Landau levels. Since the expectation of the linear statistic is exactly $\int_X f\,dv_X$, the estimator $\widehat J$ is unbiased, and its mean squared error is asymptotically $\sigma^2/N_p^{(n+1)/n}$. This extends the Bergman-kernel Monte Carlo method from compact complex manifolds to prequantized symplectic manifolds, with the spectral subspace of the Bochner–Schrödinger operator playing the role usually played by holomorphic sections.

Load-bearing premise

The construction needs the possible energy levels to have a gap, so that an interval $I$ whose endpoints avoid those levels gives a finite spectral window; if the bands overlap into a half-line there is no such window, and the paper only guarantees the gap in special geometric cases, while the variance formula also leans on an asymptotic quoted from [12].

Editorial extensions

If this is right

  • The estimator is unbiased for every $C^1$ integrand: $\mathbb E[\widehat J]=\int_X f\,dv_X$ for all $p$ large enough.
  • The mean squared error decays as $N_p^{-(n+1)/n}$, matching the optimal worst-case rate for randomized $C^1$ integration in dimension $2n$.
  • The construction works with higher Landau levels, not only the lowest one, so the resulting point processes include polyanalytic-type ensembles beyond the holomorphic Bergman case.
  • The metric $g$ and potential $V$ enter the construction, giving the user freedom to choose the auxiliary geometry while keeping the same symplectic form and volume form.
  • When the spectral window collects several Landau levels, the variance bound improves relative to a single level, because the repulsion in the determinantal process lowers fluctuations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension beyond the paper is to weaken the gap condition: a smoothed or mollified spectral window might still define a useful point process even when the Landau bands overlap, extending the estimator to all prequantized symplectic manifolds.
  • Because the variance constant in (1.13) depends on the metric $g$ and on the window $I$, one could optimize these choices for a fixed integrand, treating the geometry as a design parameter rather than data.
  • The same log-Laplace mechanism should yield central limit theorems for DPPs built from other spectral projectors, for instance generalized Bergman kernels for holomorphic vector bundles, where Slater determinants remain the natural joint density.
  • A numerical check on $S^2$ or on a flat torus with constant magnetic field, comparing single-level and multi-level windows, would show whether the predicted $N_p^{-(n+1)/n}$ rate and variance constants appear already at moderate $p$.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The paper proposes an unbiased Monte Carlo estimator for the integral of a C^1 function on a compact prequantized symplectic manifold (X, B, g) against a smooth Riemannian volume form, using determinantal point processes built from spectral projections P_{p,I} of the Bochner-Schrödinger operator H_p = (1/p)\Delta^{L_p} + V. Under the assumption that the Landau-level set \Sigma has a gap and that I is a gap interval, the spectral subspace H_p = Im P_{p,I} is finite-dimensional and defines a DPP. Theorem 1.4 asserts a central limit theorem for the normalized linear statistics \Xi_p, with variance (1.13) expressed through |df_B|^2_I, and consequently a mean squared error of order \sigma^2 / N_p^{(n+1)/n}, matching Bakhvalov's optimal worst-case rate. The proof follows the Berman and Lemoine-Bardenet strategy, replacing Bergman kernel estimates by semiclassical estimates for Bochner-Schrödinger operators from [11] and, crucially, importing the variance asymptotic (2.24) from the author's to-appear paper [12]. Examples include the almost-Kähler case with V=0 and the Guillemin-Uribe renormalized Laplacian.

Significance. If the imported asymptotic (2.24) is correct, Theorem 1.4 is a meaningful extension: it replaces holomorphic sections in [14] by spectral subspaces of Bochner-Schrödinger operators, so it applies to symplectic, not necessarily Kähler, manifolds and includes polyanalytic and higher-Landau-level processes. The paper's main structure is appealing: unbiasedness is proved directly in (2.11), the variance upper bound (1.14) is derived in the text, the spectral-gap assumption is explicitly stated with two nontrivial families of examples, and the CLT follows from a standard Montel argument once the second derivative of the log-Laplace transform converges. The paper's main weakness is self-containment: the central variance estimate (2.24), the well-definedness of |df|^2_I, and several spectral estimates are quoted from the author's earlier work or from a to-appear preprint, so the theorem as presented is conditional on external analytic results.

major comments (3)
  1. [Section 2, Eq. (2.24)] The asymptotic (2.24) is the key quantitative input of the proof: together with (2.23) it yields (2.25), and hence the variance limit (2.26) and the CLT. The sentence 'By [12, Proof of Theorem 1.2]' is the only derivation offered, and [12] is a to-appear preprint by the author. This is load-bearing and cannot be checked from the manuscript. The paper should either reproduce the proof of (2.24) in an appendix or state Theorem 1.4 as conditional on the companion paper and include enough detail for a referee to verify the constants and the multi-level case I.
  2. [Section 2, Eq. (2.17)] The proof of the commutator estimate (2.17) is reduced to 'Using this identity and the estimates (2.15) and (2.19), we can easily complete the proof'. This estimate is needed to pass from the t-dependent Hilbert-Schmidt norm to the t=0 norm in (2.25), so it participates in the uniform-on-compacts convergence (2.26). The reduction is not immediate: the displayed formula for [P_{p,I,t}, f_p] contains three terms involving products with e^{-tu_p}-1 and inverses, and the O(t p^{-(n-1)/2}) bound must be uniform in p and t. Please provide the full argument or a detailed sketch with all norm estimates.
  3. [Section 1, Eq. (1.10)] The statement that x \mapsto |df(x)|^2_I is a well-defined continuous function is deferred to [12, Section 4]. This matters for Theorem 1.4 because (1.13) and (1.14) require this object to be a genuine squared gradient: in particular, the quadratic form with coefficients \alpha_m should be nonnegative definite for every finite K_I, and the paper does not prove this. Since [12] is to-appear, this point should be settled here, at least by stating the relevant proposition from [12] with its hypotheses.
minor comments (3)
  1. [Section 2, display (2.16)] In the first equality of (2.16), the bracket should be [P_{p,I,t}, f_p], not [P_{p,I,t}, p].
  2. [Abstract and Section 1] The abstract and the introduction do not state the standing spectral-gap assumption on \Sigma until after Theorem 1.1; because the construction of H_p and the DPP requires a gap interval I=(\alpha,\beta) with \alpha,\beta \notin \Sigma, this hypothesis should be announced in the abstract or at the beginning of the introduction.
  3. [References [9] and [18]] References [9] and [18] both list arXiv:2308.04825; the second entry appears to have the wrong arXiv identifier and should be checked.

Circularity Check

1 steps flagged · score 6.0 of 10

Theorem 1.4's variance formula is imported from the author's to-appear companion [12] via equation (2.24), making the central quantitative claim dependent on an unexhibited self-cited asymptotic.

  1. self citation load bearing [Section 2, equation (2.24), feeding (2.25)-(2.26) and hence variance (1.13) in Theorem 1.4]
    "By [12, Proof of Theorem 1.2], we know that (2.24) ∫X∫X |P_{p,I}(x,y)|^2_{h_{L^p}} (f_B(x)-f_B(y))^2 dv_X(x)dv_X(y) = (1/(2π)) p^{n-1}/(2π)^{n-1} ∫X |d f_B(x)|^2_I Ω_B(x) + o(p^{n-1}), p→∞."

    This is the core variance estimate. The paper combines it with (2.23), (2.21), and (2.12) to obtain the limit (2.26) of d²F_p/dt², and hence the variance σ² in (1.13) and the claimed MSE rate. The asymptotic is not proved in the present paper; it is quoted from [12], a to-appear companion by the same author. The regularity of the variance integrand |d f_B|²_I is likewise referred to [12, Section 4]. Thus the quantitative statement of Theorem 1.4 reduces to an unexhibited self-cited result: if (2.24) failed for a multi-level interval I, the stated variance and the optimal-rate constant would collapse. The qualitative CLT mechanism is independently supported by Berman and Lemoine-Bardenet, but the specific variance formula is not independently checkable from this text.

full rationale

The paper is not globally circular: the determinantal point process is explicitly constructed via Slater determinants, the estimator's unbiasedness is an exact identity from the DPP linear-statistics formula, and the CLT argument follows the Berman/Lemoine-Bardenet scheme. However, the genuinely new quantitative output, the asymptotic variance σ² in (1.13), is obtained through the imported asymptotic (2.24), which is quoted as known from [12, Proof of Theorem 1.2], a to-appear companion by the same author; the paper also refers to [12, Section 4] for the well-definedness of |d f|²_I. No derivation of this lemma is included here. This is a load-bearing self-citation: the central claim reduces, at the level of the variance constant, to [12]. I find no fitted-input-called-prediction, no uniqueness-imported-from-authors, and no renaming of a known result. Spectral estimates from [11] are published analytic input rather than a substitute for the theorem, so they do not by themselves raise the score. Because the central variance formula depends on an unexhibited self-cited asymptotic, the score is 6.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central theorem rests on the prequantization condition, a spectral gap, and imported same-author estimates from [11,12]. There are no data-fitted numerical parameters; the potential V and the interval I are design choices that the theorem treats as given, not fitted values.

assumptions (5)
  • domain assumption Prequantization: [B] is in H^2(X,2pi Z), so there exists a Hermitian line bundle (L,h^L) with connection grad^L such that B=iR^L (Eq. 1.1).
    Restricts the class of symplectic manifolds to prequantized ones; without this condition no such line bundle exists.
  • domain assumption Spectral gap: the Landau-level set Sigma has a gap and an interval I=(alpha,beta) with alpha,beta not in Sigma is chosen (text before Theorem 1.4).
    Ensures P_{p,I} has finite rank for large p. For generic (X,B,g,V), Sigma may be a half-line, so this is restrictive.
  • domain assumption Smooth geometric data: dv_X is the Riemannian volume of a smooth metric g, and the Bochner-Schrodinger operator H_p uses a smooth potential V.
    The spectral projection and its kernel asymptotic require smoothness of the metric, the volume form, and the potential.
  • ad hoc to paper Imported estimates from [11] and [12]: the spectral gap description (Theorem 1.1), kernel expansion (2.20), exponential decay of the kernel, and the variance asymptotic (2.24).
    The proof of Theorem 1.4 imports these estimates without derivation; they are load-bearing and come from same-author references.
  • domain assumption The function f is C^1 and f_B = f / sqrt(det B_x) is integrated against the Liouville volume form.
    The central limit theorem and variance formula are stated for such functions, and the proof uses the C1 regularity through the variance asymptotics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Monte Carlo methods on compact symplectic manifolds." pith.science (2026). https://pith.science/paper/ZVW2DDP4

@misc{pith2026260807021,
  author       = {Pith},
  title        = {Pith review of: Monte Carlo methods on compact symplectic manifolds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZVW2DDP4}},
  note         = {Machine review of arXiv:2608.07021}
}
abstract

We build an unbiased Monte Carlo estimator of the integral of any $C^1$ function on a prequantized compact symplectic manifold against a smooth Riemannian volume form, taking for quadrature nodes the determinantal point process associated with an appropriate spectral projection of the Bochner-Schr\"odinger operator. We show that the estimator satisfies a central limit theorem, and the decay rate of the mean squared error reaches the optimal worst-case rate investigated by Bakhvalov in Euclidean spaces. These results extend previous results of Lemoine and Bardenet on Monte Carlo methods on compact complex manifolds.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 7 canonical work pages

  1. [12]

    Yu. A. Kordyukov, Determinantal point processes associated with the Bochner- Schr¨ odinger operator, to appear inRuss. J. Math. Phys.33(2026), preprint arXiv:2605.13575 (2026)

    work page Pith review arXiv 2026

  2. [11]

    Yu. A. Kordyukov, Semiclassical spectral analysis of the Bochner-Schr¨ odinger opera- tor on symplectic manifolds of bounded geometry,Anal. Math. Phys.12(2022). no. 1, Paper No 22, 37 pp

    2022

  3. [14]

    Monte Carlo methods on compact complex manifolds using Bergman kernels

    T. Lemoine and R. Bardenet. Monte Carlo methods on compact complex manifolds using Bergman kernels. preprint arXiv:2405.09203, 2024

    work page Pith review arXiv 2024

  4. [1]

    N. S. Bakhvalov. Approximate computation of multiple integrals. (Russian)Vestnik Moskov. Univ. Ser. Mat. Meh. Astr. Fiz. Him.1959(1959), no, 4, 3–18; English translation: On the approximate calculation of multiple integrals,J. Complexity31 (2015), 502–516

    work page 1959

  5. [2]

    Monte Carlo with determinantal point processes.Ann

    Bardenet, R.; Hardy, A. Monte Carlo with determinantal point processes.Ann. Appl. Probab.30 (2020), no. 1, 368–417

    work page 2020

  6. [3]

    Berman, Determinantal point processes and fermions on complex manifolds: large deviations and bosonization.Comm

    R.J. Berman, Determinantal point processes and fermions on complex manifolds: large deviations and bosonization.Comm. Math. Phys.327(2014), 1–47

    2014

  7. [4]

    Berman, Determinantal point processes and fermions on polarized complex man- ifolds: bulk universality, InAlgebraic and analytic microlocal analysis,Springer Proc

    R.J. Berman, Determinantal point processes and fermions on polarized complex man- ifolds: bulk universality, InAlgebraic and analytic microlocal analysis,Springer Proc. Math. Stat., vol. 269, Springer, Cham, 2018, pp. 341–393

    2018

  8. [5]

    Charles, On the spectrum of nondegenerate magnetic Laplacians.Anal

    L. Charles, On the spectrum of nondegenerate magnetic Laplacians.Anal. PDE17 (2024), no. 6, 1907–1952

    2024

Show all 17 references
  1. [6]

    Gautier, R

    G. Gautier, R. Bardenet, M. Valko, On Two Ways to Use Determinantal Point Pro- cesses for Monte Carlo Integration. InAdvances in Neural Information Processing Systems (NeurIPS),volume 32. 2019. Curran Associates, Inc

    2019

  2. [7]

    Guillemin and A

    V. Guillemin and A. Uribe, The Laplace operator on thenth tensor power of a line bundle: eigenvalues which are uniformly bounded inn.Asymptotic Anal.1(1988), 105–113

    1988

  3. [8]

    Haimi and A

    A. Haimi and A. Wennman, A central limit theorem for fluctuations in Polyanalytic Ginibre ensembles.Int. Math. Res. Not. IMRN,2019, no. 5, 1350–1372

    2019

  4. [10]

    Yu. A. Kordyukov, On asymptotic expansions of generalized Bergman kernels on symplectic manifolds. (Russian); translated fromAlgebra i Analiz30(2018), no. 2, 163–187St. Petersburg Math. J.30(2019), no. 2, 267–283 16 Y. A. KORDYUKOV

    2018

  5. [13]

    T. Lemoine. Determinantal point processes associated with Bergman kernels: con- struction and limit theorems, preprint arXiv:2211.06955 (2022)

    2022 arXiv

  6. [15]

    Ma and G

    X. Ma and G. Marinescu,Holomorphic Morse inequalities and Bergman kernels. Progress in Mathematics, 254. Birkh¨ auser Verlag, Basel, 2007

    2007

  7. [16]

    Ma and G

    X. Ma and G. Marinescu, Generalized Bergman kernels on symplectic manifolds.Adv. Math.217(2008), 1756–1815

    2008

  8. [17]

    E. Novak. Some results on the complexity of numerical integration. Monte Carlo and quasi-Monte Carlo methods, 161–183,Springer Proc. Math. Stat.,163, Springer, [Cham], 2016

    2016

  9. [18]

    Petrovic, R

    V. Petrovic, R. Bardenet, A. Desolneux, Repulsive Monte Carlo on the sphere for the sliced Wasserstein distance,Transactions on Machine Learning Research Journal,In press; Preprint arXiv:2308.04825. Institute of Mathematics, Ufa Federal Research Centre, Russian Academy of Scie...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.